Benchmarking Database Storage

Recently a number of disk-based storage array vendors and hybrid array vendors have been posting TPC-C results they achieved with the HammerOra tool.  Here’s our take on the subject.


Here at Pure Storage, our TPC-C tests have been pretty impressive.  We regularly run our unaudited HammerOra harness at over 1.73 million tpm (in ARCHIVELOG mode!), which ranks 6th among the audited results.

 

The problem with HammerOra is that it doesn’t actually push the storage sub-system to its limits.  The bottleneck in these tests is almost always a combination of CPU resources and concurrency.  Maybe that’s why you don’t see any storage vendors sponsoring TPC benchmarks.

To demonstrate what our storage is actually capable of, we like to use the “Silly Little Oracle Benchmark”  (SLOB) tool which was developed by Kevin Closson.  Kevin was the Performance Architect in Oracle’s Exadata group, and he now works for EMC.  Unlike HammerOra (and Swingbench), SLOB is specifically designed to drive a lot of physical I/O without any application contention.  In other words, SLOB shows just how much Oracle-generated I/O your storage subsystem can do.

For example, on the read-intensive test (128 readers, 0 writers), the Pure Storage FlashArray can drive over 2GB/s of throughput at over 250,000 IOPS with sub-millisecond latency:

Naturally, latency is a concern.  As the histogram shows, we average 1ms latency for the duration of the test run.

We will have more SLOB results and discussions soon.

If you are evaluating a Flash Storage solution, send us your AWR report and we’ll schedule a free consultation with our Oracle performance experts to talk about how Pure Storage’s FlashArray can improve the performance of your Oracle environment while actually reducing the cost of your storage architecture.